Fish food web bioaccumualtion model:

Goal:

To evaluate the model performance characterizing and predicting values in three samples systems (Joint Base Andrews, JBA, in spring and summer, and Willow Grove). In a given system, each species is described by the mean body size of all collected fish, and the system is described by the mean measured temperature, dissolved oxygen levels, organic carbon when available. No invertebrates were found or sampled in these systems.

The model:

The food web model is adapted from Sun et al 2020.

Metrics of model performance:

  1. \(R^2\) (full model, by PFAA, by Species)

\(R^2 = 1-RSS / TSS\)

$RSS = (modeled_{val} - observed_{val}) $

$TSS = (modeled_{val} - mean(observed_{val})) $

  • RSS = residual sum squared
  • TSS = total sum squared
  1. Model bias (MB)

The method from: McLeod et al 2016.

\(MB_{avg} = (log(C_{avg-pred})/log(C_{avg-model}))\)

\(MB_{CV} = (log(C_{CV-pred})/log(C_{CV-model}))\)

  • C[avg-model] = average of predicted PFAA concentration for all organisms and all PFAS
  • C[avg-obs] = average of observed PFAA concentration for all organisms and all PFAS
  • CV = coefficients of variation (CV)


1. Willow Grove

The model predictions for Willow Grove. This system has 3 fish species, and was sampled in fall.

Figure. 1

Figure. 1

PFAS predictions in Willow Grove system. (A) The PFAS in each diet item were known and provided during the modelling. (B) The PFAS values were estimated based on the food web trophic interactions, the only known PFAS values here were in water and sediment. The dot models values when fish tissue PFAS, water, and sediment values are at their measured mean values. The food web structure was the same between both cases. The dotted lines mark values where data were over- or under-predicted by a factor or 2 (lower line) and 10 (upper line), respectively.



2. Joint Base Andrews (spring)

The model predictions for JBA in spring season. This system has 8 fish species, and was samples in spring.

Figure. 2

Figure. 2

PFAS predictions in JBA system in spring season. (A) The PFAS in each diet item were known and provided during the modelling. (B) The PFAS values were estimated based on the food web trophic interactions, the only known PFAS values here were in water and sediment. The dot models values when fish tissue PFAS, water, and sediment values are at their measured mean values. The food web structure was the same between both cases. The dotted lines mark values where data were over- or under-predicted by a factor or 2 (lower line) and 10 (upper line), respectively.



3. Joint Base Andrews (summer)

Figure 3

Figure 3

PFAS predictions in JBA system in summer season. (A) The PFAS in each diet item were known and provided during the modelling. (B) The PFAS values were estimated based on the food web trophic interactions, the only known PFAS values here were in water and sediment. The dot models values when fish tissue PFAS, water, and sediment values are at their measured mean values. The food web structure was the same between both cases. The dotted lines mark values where data were over- or under-predicted by a factor or 2 (lower line) and 10 (upper line), respectively.



All systems together

Figure 4

Figure 4

PFAS predictions in all systems together. (A) The PFAS in each diet item were known and provided during the modelling. (B) The PFAS values were estimated based on the food web trophic interactions, the only known PFAS values here were in water and sediment. The dot models values when fish tissue PFAS, water, and sediment values are at their measured mean values. The food web structure was the same between both cases. The dotted lines mark log values of 2 and 10 showing where values are over- or under-predicted by a factor or 2 and 10 respectively.



Error estimates and model bias

Table 1: Average model bias
Model Data type JBA-summer JBA-spring Willow Grove
known diet median values 1.03 1.08 1.04
trophic transfer median values 1.04 1.10 1.14
known diet same day values 0.99 1.08 1.03
trophic transfer same day values 0.99 1.10 1.13
Table 2: Average model bias by PFAA
System Model Data type PFDA PFHxS PFNA PFOA PFOS PFUA
Willow Grove known diet median values NaN 1.09 0.80 0.77 1.03 NaN
JBA-summer known diet median values 1.15 1.11 0.95 0.98 1.03 NA
JBA-spring known diet median values 1.17 1.11 1.04 0.94 1.07 NA
Willow Grove trophic transfer median values NaN 1.12 0.74 0.76 1.12 NaN
JBA-summer trophic transfer median values 1.30 1.12 0.94 0.99 1.03 NA
JBA-spring trophic transfer median values 1.32 1.12 1.05 0.94 1.09 NA
Willow Grove known diet same day NaN 1.04 0.80 0.72 1.03 NaN
JBA-summer known diet same day 1.15 1.03 0.91 0.90 0.99 NA
JBA-spring known diet same day 1.18 1.11 1.04 0.94 1.07 NA
Willow Grove trophic transfer same day NaN 1.06 0.73 0.71 1.11 NaN
JBA-summer trophic transfer same day 1.30 1.03 0.89 0.90 0.99 NA
JBA-spring trophic transfer same day 1.32 1.12 1.06 0.94 1.08 NA



Figure 5.

Figure 5.

Sample size of observed values and species specific error

Figure 6.

Figure 6.

Figure 7.

Figure 7.

Figure 8.

Figure 8.

The relationship between the number of individuals collected at a given sampling event and the difference between the mean observed tissue concentrations and model predicted concentrations.



Contributions to PFAA bioaccumulation via diet and water


Figure 9.

Figure 9.

Figure 10.

Figure 10.

The predicted % contribution from gill uptake and dietary uptake in total PFAS uptake in each fish species in each system. JBA sp = Joint Base Andrews, spring season. JBA sum = Joint Base Andrews, summer season. WG = Willow Grove. Species Abbreviations:


The modeled uptake rates and elimination of PFAS by each species.


Figure 11.

Figure 11.

The % uptake that come from diet is positively correlated with total PFAS tissue bioaccumulation. Each dot is a species.


Figure 12.

Figure 12.

The chain length of perfluoroalkyl carboxylic acids (PFCA) and perfluorosulfonic acids (PFSA) and the % diet contribution in total PFAS uptake. Each dot is species; all systems are presented together.


Figure 13.

Figure 13.

The chain length of perfluoroalkyl carboxylic acids (PFCA) and perfluorosulfonic acids (PFSA) and their tissue bioaccumulation. Each dot is species; all systems are presented together. The open circles are observed values, the closed circles are modeled values.


Predicted relationships between sediment:water PFAS ratio and fish tissue bioconcentrations (A), and predicted diet uptake (B). Each dot is a species.

The proportion of PFAS predicted uptake just from the water

Figure 14.

Figure 14.

Figure 15.

Figure 15.

Dietary PFAS uptake (%) is higher in fish at higher trophic levels.